KR101346965B1 - Image processing device, measuring/testing system, and recording medium - Google Patents

Abstract

The present invention provides an outline extraction technique capable of eliminating not only excess, overlap or whiskers of the outline of the sample pattern, but also breaks in the outline. The thinning process is performed on the design data for generating the sample pattern, and pattern internal and external specification information defining the inside and the outside of the pattern formed on the target sample is generated. Based on the Marker Controlled Watershed Segmentation method, region division is performed by expanding the region represented by the prescribed information inside and outside the pattern while referring to the pixel value of the edge-enhanced image of the target sample image to generate a pattern outline.

Description

Image processing unit, measurement / inspection system and recording medium {IMAGE PROCESSING DEVICE, MEASURING / TESTING SYSTEM, AND RECORDING MEDIUM}

TECHNICAL FIELD The present invention relates to an image processing apparatus, a measurement / inspection system, and a program, and more particularly, to a technique for performing contour extraction of a pattern on a sample.

Background Art Conventionally, a technique for extracting contour lines from a target image such as an SEM image obtained by photographing a semiconductor pattern is called edge extraction or contour extraction and is widely used. For example, Patent Literature 1 discloses a technique of forming edges of a pattern by emphasizing edges using a nonlinear second derivative filter, binarizing and thinning, extracting edges, combining the extracted edges, and approximating a straight line. Is disclosed. In addition to the second derivative, a first derivative filter such as a Sobel filter may be used for edge emphasis.

However, the method of extracting the edge from such local information has a problem that breaks, surpluses, overlaps or whiskers appear in the middle of the outline of the semiconductor pattern.

On the other hand, the technique of searching for the shape registered in advance from the contrast image is widely used as a template matching. For example, Patent Document 2 describes a design data and a semiconductor pattern, which are basic data for generating a semiconductor pattern, are aligned by template matching, and an edge extracted by a filter of a first derivative or a second derivative is designed. A method of forming an outline of a pattern in association with data is disclosed. According to this method, it is possible to eliminate the excess, overlap or whisker of the edge in the outline.

Japanese Patent Application Publication No. 2006-66478 Japanese Patent Application Laid-Open No. 2001-338304

IEEE transactions on Pattern Analysis and machine Intelligence, 25 (3): 330-342, March 2003, "Water snakes Energy driven Watershed segmentation," by H. T. Nguyen, M. Worring and R. V. D Boomgard International Journal of Computer Vision, v. 82 n.3, p.264-283, May 2009, "Four-Color Theorem and Level Set Methods for Watershed Segmentation," by Erlend Hodneland, Xue-Cheng Tai, Hans-Hermann Gerdes

However, when the technique disclosed in Patent Literature 2 is used, the problem of surplus, overlap, and whisker of edges in the contour of the pattern is solved, but the interruption in the middle of the contour may still not be solved.

This invention is made | formed in view of such a situation, and it is providing the technique which can eliminate not only the excess, overlap, or whisker of the outline of a sample pattern, but also the interruption of an outline.

In order to solve the said subject, in this invention, the thinning process is performed with respect to the design data for producing | generating a sample to be examined, and the pattern internal and external specification information which prescribes the inside and the outside of the pattern formed on the target sample is produced | generated. Then, while referring to the pixel values of the edge-enhanced image of the target sample image, area division is performed by expanding the region represented by the pattern internal and external prescribed information to generate a pattern outline of the target sample.

When generating the outline, the area indicated by the internal and external pattern information is expanded so that the area inside the pattern becomes the edge portion indicated by the edge emphasis image (the obtained edge emphasis image may be broken in the middle). Since it does not overlap%, area | region division is made so that it may overlap. For example, region division is performed based on the Marker Controlled Watershed Segmentation method. Alternatively, the pattern contour of the target sample is generated by performing dynamic contour extraction processing using the external energy representing the pattern inner region and the external energy representing the pattern outer region from the pattern internal and external prescribed information. Examples of this dynamic contour extraction processing include a snake method and a level set method.

When the pattern internal and external prescribed information is generated, a peeling process is performed on the image of the inner region and the outer region of the pattern in the design data with reference to the position alignment information obtained by matching the target sample image with the design data. In this case, the design data may be modified so that the design data includes the target sample image, and the prescribed information inside and outside the pattern may be generated from the modified design data.

In addition, when generating an edge emphasis image, you may perform a some filter process with respect to a sample image, and may combine each filter process image.

Further features of the present invention will become apparent from the best mode for carrying out the present invention and the accompanying drawings.

According to the process of the present invention, not only the excess, overlap or whisker of the outline of the sample pattern, but also the interruption of the outline can be eliminated.

1 is a diagram illustrating a schematic configuration of a measurement / inspection system according to an embodiment of the present invention.
2 is a diagram illustrating an example of design data.
3 is a diagram illustrating an example of an SEM image of a semiconductor pattern.
4 is a diagram illustrating an example of pattern information in and out of the pattern.
It is a figure which shows the state which superimposed pattern internal and external definition information on the SEM image.
FIG. 6 is a diagram illustrating edge emphasis images generated from SEM images. FIG.
FIG. 7 is a diagram showing pattern internal and external separation information generated from pattern internal and external definition information.
Fig. 8 is a diagram showing an SEM outline generated by the process of the present invention.
9 is a diagram illustrating a state where an SEM outline and an SEM image are superimposed.
10 is a view for explaining the watershed segmentation method.
11 is a view for explaining the Marker Controlled Watershed Segmentation (MCWS) method.
It is a figure which shows the internal structure of the pattern internal and external regulation information generation part which concerns on 2nd Embodiment of this invention.
It is a figure for demonstrating the concept of the pattern internal and external regulation information generation process which concerns on 2nd Embodiment.
FIG. 14 is a diagram for explaining the processing of the pattern internal and external regulation information generation unit 102 when a plurality of patterns are imprinted on the SEM image.
It is a figure which shows schematic structure of the edge emphasis part which concerns on 3rd Embodiment of this invention.
FIG. 16 is a diagram (1) for explaining the concept of a connection cost with a marker in MCWS.
17 is a diagram (2) for explaining the concept of a connection cost with a marker in MCWS.
It is a figure for demonstrating the snake method.
It is a figure for demonstrating the level set method.

According to the present invention, whiskers are generated by generating pattern definition information defining the inside and the outside of a pattern from an acquired image (for example, an SEM image) and design data by an electron microscope, and performing segmentation processing using the edge image. It relates to a technique for generating pattern contours, overlapping and no interruption.

EMBODIMENT OF THE INVENTION Hereinafter, embodiment of this invention is described with reference to an accompanying drawing. It should be noted, however, that the present embodiment is merely an example for realizing the present invention and does not limit the technical scope of the present invention. In addition, the same reference number is attached | subjected about the structure common in each figure.

(1) First Embodiment

<Configuration of Measurement / Inspection Device>

1 is a diagram showing a schematic configuration of a measurement / inspection system (also referred to as an "image observation system") 1 according to an embodiment of the present invention. In the present embodiment, the image processing using the SEM image of the semiconductor pattern has been described, but any sample may be used as long as it is not limited to the semiconductor pattern and can be an object of observation.

The measurement / inspection system 1 includes an image processing unit 10, an electron microscope body unit 20, an SEM image generating unit 30, a control calculator 40, an input unit 50, and a display device ( 60). In the electron microscope main body 20, the electron beam 202 emitted from the electron gun 201 is converged by an electron lens (not shown) and irradiated to the sample (for example, a semiconductor chip) 204. By the electron beam irradiation, the intensity of the secondary electrons or the reflected electrons generated from the sample surface is detected by the electron detector 205 and amplified by the amplifier 206.

The deflector 203 which moves the position of an electron beam is controlled by the control signal 207 of the control calculator 40, and raster scans the electron beam 202 on the sample surface.

The amplified data is provided to the SEM image generating unit 30 to generate a SEM image. This generated SEM image is provided to the image processing unit 10, and predetermined image processing is executed. The SEM image is displayed on the display device 60 and stored in an image memory (not shown).

The image processing unit 10 performs matching (position alignment) between the SEM image and the design data to generate position alignment information, and defines the inside and the outside of the pattern in the semiconductor pattern from the position alignment information. A pattern internal and external regulation information generation unit 102 for generating pattern internal and external regulation information (which indicates which part is inside and outside), an edge emphasis unit 103 for generating an edge emphasis image by emphasizing the edge of the SEM image, A pattern inner and outer separating unit 104 that separates the inside and the outside of the pattern from the edge emphasis image and the pattern inner and outer defining information, and outputs the inner and outer pattern separating information, and the contour generating unit 105 which generates the SEM contour from the inner and outer pattern separating information. Equipped with.

The input unit 50 is an input means such as a keyboard or a mouse, and inputs a current value or a voltage value for controlling the electron microscope main body 20, or inputs design data of a semiconductor pattern obtained by obtaining an SEM image by an electron microscope. It is a tool to do it. In addition, regarding the design data, the data stored in the image memory may be read.

<Output Information of Each Part of the Image Processing Unit 10>

2 shows an example of design data. For convenience, the white part is set inside the pattern and the black part is set outside the pattern. 3 shows an SEM image of a semiconductor pattern created from this design data.

4 shows corresponding information in and out of the corresponding pattern. FIG. 4A is obtained by performing thinning processing on the white region (for example, inside the pattern) and the black portion (outside the pattern) shown in FIG. 2. In other words, the prescribed information inside and outside the pattern is a prescribed number of peeling (thinning) processes (though one pixel of the outer periphery is cut out (removed) for the image in which the respective areas of the inside and the outside are painted in the design data. The pixel part is created by processing which is not cut off. 4 (b) is a diagram showing simplified information inside and outside the pattern shown in FIG. 4 (a) in order to make it easy to see FIG. 4 (a).

Since the pattern information is created in consideration of the design data and the position alignment information of the SEM image, as shown in FIG. 5, when the display is superimposed on the SEM image, the brightest region defines the inside of the pattern, and then It can be seen that the bright areas define the outside of the pattern. Therefore, it can be seen that a boundary (edge portion) between the inside and the outside of the pattern exists between the inside pattern information and the outside pattern information.

6 shows an edge emphasis image created by filtering the SEM image. 7 shows separation information inside and outside the pattern. The separated information inside and outside the pattern is each inside (lightest part) and outside of FIG. 4 so that the part with the high luminance value of the edge-enhanced image becomes a boundary between the inside and outside (darkest part) defined by the inside and outside pattern prescribed information. It is information obtained by dividing a line indicating (the next brightest portion) (swelling each line until the inside and the outside are connected), and separating the entire image into the inside and the outside of the pattern. Details of this processing will be described later.

8 shows an SEM outline created by the outline creating unit 105 as a boundary between the inside and the outside of the pattern inside and outside separation information. 9 shows that the SEM outline of FIG. 8 is superimposed on the SEM image of FIG. 3. As described above, it can be seen that the SEM contour is surely obtained.

<Details of the process of separating part inside and outside the pattern>

The portion (internal and external pattern separation) for generating the internal and external pattern separation information from the pattern internal and external regulation information and the edge-enhanced image is processed in accordance with the Marker Controlled Watershed Segmentation (MCWS). First, prior to the description of the MCWS, the Watershed Segmentation which is the basis thereof will be described.

(Ⅰ) Watershed Segmentation

In Watershed Segmentation, region division is performed on an edge-enhanced image in which pixel values indicate that the edge of each pixel of the original image (for example, an SEM image) is expressed. Considering the edge-enhanced image as the topographical information (regarding the luminance value of the pixel as the height), the edge portion is like a branch partitioned by a high mountain range and other edge mountain ranges. For such terrain, if the water level (corresponding to the luminance value of the pixel) is submerged while being kept the same throughout the image, a large number of ponds are created each time the water level rises, and they are connected, and finally the entire terrain is flooded. do. Each time a new pond is created in this process, a different label is given to the pond, and pixels belonging to each pond are classified into areas specified by the label of the pond. As the water level rises, the pond widens and one day connects with the neighboring pond, but its connected location becomes the boundary of the area. The boundary of the area becomes the shape connecting the peaks of the mountain range. Since these peaks divide the pond, this method of segmentation is called Watershed.

10 is a diagram conceptually showing an example of an algorithm for implementing watershed segmentation. In FIG. 10, the thick dotted line shows the submerged line (intersection between the surface and the terrain) at a certain level. The water level moves from the height of the lowest point of the terrain to the height of the highest point, but the following three steps are performed on all the pixels on the submerged line at each level.

STEP1: If a neighboring pixel is a pond, the neighboring pond is also labeled.

STEP2: Enlarge the same label to adjacent pixels on the submerged line.

STEP3: In STEP1 and 2, a new label is given to non-target pixels (bottom of basin, new pond). In order to imagine the water level rising in watershed segmentation, the surface type is `` slowly immersed in the bath in the bathtub '' or `` a small hole in the bottom of the basin '' rather than the `` rain and water. '' Is slowly immersed in a tub of uneven plastic plate. This is because there is no guarantee that the water level will be constant by rain. Since watershed segmentation performs finer area division than necessary, it cannot be applied to the creation of outlines as it is.

(Ii) Marker Controlled Watershed Segmentation (MCWS)

Based on the knowledge of ordinary watershed segmentation, MCWS is demonstrated continuously. MCWS improves the watershed segmentation to control the method of segmentation by markers. When the water level rises in watershed segmentation as "slowly immersing the sponge topography in the bathtub," the MCWS only uses a sponge (corresponding to a segment of internal and external regulatory information). It is conceivable to consider a situation in which a member (for example, plastic) does not pass through. In addition, when the water level rises in the watershed segmentation, the surface of the basin is roughly immersed in the tub with a surface-shaped uneven plastic plate with a small hole in the basin. In other cases (even at the bottom of the basin), it is conceivable that there is no hole.

By setting the markers in this way, the number of regions to be finally divided coincides with the number of markers so that all the regions contain markers. The shape of the marker may be a dot, a line or a face. In addition, although a label is provided for every marker, you may give the same label to some marker. In this case, even if ponds of the same label are connected, it is not recognized as a boundary, and the number of areas is less than the number of markers and more than the number of labels.

11 is a diagram for conceptually explaining an example of an algorithm for realizing MCWS. The viewpoint of drawing is the same as FIG. The following three steps are performed at each water level (not all the pixels on the flooded line).

STEP1: The pixel on the marker is given the same label as the marker.

STEP2: If the neighboring pixel is a pond, the neighboring pond is also labeled.

STEP3: Enlarge the same label on adjacent pixels that are above the immersion line or below the water level and have no label.

Compared with normal watershed segmentation, it is characteristic that MCWS does not give new label, and the order of STEP changes in order to take over the label of marker (1, 2, 3 → 3, 1, 2), and MCWS The label is expanded in STEP3 (corresponding to STEP2 of normal watershed segmentation) for pixels whose height is below the level, that is, flooding of the pond occurs. Branches without markers are labeled by flooding of adjacent ponds through the peaks, but flooding occurs from the pond facing at the lowest point among the peaks surrounding the basin.

In addition, the pattern information of the inside and outside of the pattern of FIG. 4 is a marker in MCWS, and the brightest area | region and the next bright area | region show the marker area from which a label differs. In addition, the edge emphasis image of FIG. 5 corresponds to the terrain information. 7 shows a state in which the entire image is divided into two regions, namely the inside of the pattern and the outside of the pattern, by MCWS using the inside and outside pattern defining information as markers.

Thus, since outlines (borders between the inside and outside of the pattern) are obtained from the internal and external prescribed information of the pattern, even if the edge-enhanced image may be partially broken in the middle, the resulting outline is not broken in the middle.

(2) Second Embodiment

2nd Embodiment is related with the other form of the pattern information definition part inside and outside. Other configurations and operations are the same as those in the first embodiment, and description thereof will be omitted.

12 is a diagram showing another embodiment of the pattern internal and external regulation information generation unit 102 according to the second embodiment of the present invention. As shown in FIG. 12, the pattern internal and external definition information generation unit 102 is constituted by a design data shift unit 1021, a design data deform unit 1022, and a pattern contraction unit 1023.

The design data shift unit 1021 shifts the design data to a position where the SEM image and the pattern overlap based on the position alignment information (overlap processing). Next, the design data deformable portion 1022 deforms the shifted design data so that the shape of the design data approaches the pattern shape of the SEM image (for example, the design data is blunted by the low pass filter). Finally, the pattern contraction section 1023 fills the deformed design data to generate the pattern information (internalization process).

FIG. 13 is a diagram for explaining a process executed by the internal and external prescribed information generation unit 102 shown in FIG. 12. FIG. 13A illustrates overlapping display of design data (black) and a semiconductor pattern created based on the same and observed by SEM. In general, the semiconductor pattern has a rounded shape with an angle drop compared to the design data.

From (b) of FIG. 13, the shape generated by the peeling from the design data of (a) of FIG. 13 is gathered. When overlapping the semiconductor pattern as shown in (c) of FIG. 13, the portions protrude outward. Able to know. Therefore, Fig. 13D shows that the angle of the angular design data is rounded to approximate the shape of the semiconductor pattern. As a method of rounding the design data, a method of creating a light intensity distribution image (airreal image) and obtaining a cross-sectional shape cut at the same luminance level as in the litho process simulation can be considered. A low pass filter such as a Gaussian filter with a large window and a window may be used to substitute a light intensity distribution image. In the case of a certain pattern shape by simulation, a trend such as how much the shape is distorted (rounded) may be grasped, and the design data may be rounded based on the trend.

The shape obtained by peeling this deformed design data is shown by the dotted line. Although FIG. 13E shows the semiconductor pattern superimposed thereon, it can be seen that the semiconductor pattern does not protrude outside here.

FIG. 14 is a view for explaining the processing of the pattern internal and external regulation information generation unit 102 when a pattern of a plurality of layers (here, two layers) is imprinted on the SEM image. In FIG. 14A, the lower layer pattern is imprinted on the upper and lower portions parallel to the transverse direction under the upper layer patterns of the three line ends parallel to each other. Here, when only the outline of the upper layer pattern is extracted, only the design data of the upper layer is used for processing. In general, since the edge of the upper layer is stronger than the edge of the lower layer, two kinds of markers (internal and external pattern information) can be created inside and outside of the upper layer design, and only the outline of the upper layer can be extracted without being affected by the lower edge. Can be.

However, in the case where the edge of the lower layer is strongly taken, as shown in FIG. 14B, markers outside the upper layer are disposed so as to penetrate the lower layer with reference to the design data of the lower layer (upper and other than that). Classify). In this way, both sides of the edge of the lower layer pattern boundary are recognized as the same pond in the processing of the MCWS, and the influence of the edge of the lower layer can be excluded.

The same method can be used not to exclude the influence of the lower edge, but also to exclude the influence of fake edges made during manufacturing. Since fake edges cannot be determined from the design data, it is important not to make them too thin when creating the marker by filling the design data. The degree of peeling needs to be set so that there is an error in positional alignment in consideration of the precision of positional alignment and shrinkage of the pattern, and the marker does not penetrate the pattern even if the pattern is contracted to some extent. Not desirable The marker can also be greatly deflated for a portion that tends to shrink, such as a line end. In addition, the process of scraping off a marker from design data is not limited to peeling.

In addition, when extracting not only the contour of an upper layer pattern, but also the contour of a lower layer pattern, what is necessary is just to arrange | position a marker like FIG.14 (c) using design data of a lower layer. In other words, there are three types of marker I (the brightest marker) inside the upper layer, marker II (the next brightest marker) inside the upper layer and marker III (the darkest marker) outside the upper layer.

In the SEM image in which the multilayered pattern is imprinted, there is also the one in which the lower edge is imprinted inside the upper layer. Such edges are very weak, but can be extracted by dividing the types (labels) of the markers inside the upper layer and the inside of the lower layer and the markers outside the lower layer.

(3) Third Embodiment

The third embodiment relates to another form of the edge emphasis portion. Other configurations and operations are the same as those in the first embodiment, and description thereof will be omitted.

It is a figure which shows schematic structure of the edge emphasis part which concerns on 3rd Embodiment. The edge emphasis unit 103 is composed of a primary differential generator 1031, a secondary differential generator 1032, and a synthesis unit 1033.

The primary differential generating unit 1031 creates an edge emphasis image obtained by filtering the primary differential meter on the SEM image. The secondary differential preparation unit 1032 creates an edge emphasis image obtained by filtering the secondary differential meter on the SEM image. As an example of the processing of the combining unit 1033, the second derivative is generated after matching the luminance levels of the edge-weighted image created by the first derivative generator 1031 and the edge-weighted image created by the second derivative-maker 1032. The edge emphasis image created in the section 1032 is expanded using, for example, a Maximum filter to create a maximum value image for each pixel, and finally, smoothed by a low pass filter.

The primary differential preparation unit 1031 can take an edge clearly with respect to a pattern in which the inside is bright and the outside is dark. On the other hand, the secondary differential generating unit 1032 can better take an edge with respect to the pattern in which the white band is present at the boundary although the brightness of the inside or the outside does not change. In other words, when the edge extracted by the second derivative generator 1032 is expanded, the lowering of the luminance value can be suppressed than the edge extracted by the first differential generator 1031 for the final smoothing. By this effect, when the white band is clear, the center of the white band is extracted, and when the white band is unclear, the step edge of the larger change in contrast is extracted, but the contrast change of one step edge is noticeable. The center of the white band is extracted unless it is noticeably large.

Therefore, the edge of the edge emphasis image output by the synthesis section 1033 basically takes the center of the white band of the SEM image, and when the contrast change is noticeably large on one of the inner or outer step edges, By taking a step edge, a sudden change from the inner side to the outer step edge is suppressed.

(4) Fourth Embodiment

In the fourth embodiment, in the process of the outline preparation unit 105, the energy function obtained by the MCWS is applied to the dynamic outline extraction process. Here, the snake method and the level set method are taken as examples of the dynamic contour extraction processing.

In addition to the process of creating the outline as shown in FIG. 8 as the boundary of the area of the separated information inside and outside the pattern as shown in FIG. 7 as described above, the outline creation unit 105 corrects the outline to be more natural. It is also conceivable to carry out a processing to be performed. Prior to explaining the example, the concept of the connection cost with the marker in the MCWS will be described with reference to FIGS. 16 and 17. The connection cost with the marker is a function defined at all points for the marker of each label.

In FIG. 16, there are two kinds of markers of one kind (same label). The cost of the connection with the marker is basically equal to the height of the surface at each point, but the height of the surface of the flood point (the point where the water leaked) at the point where the water sank below the surface due to the flood during the MCWS. That is, it becomes equal to the water level after flooding.

In addition, in FIG. 17, markers of a different kind (label) exist in two places. At each point, both the connection cost with marker 1 and the connection cost with marker 2 are defined. The area including marker 1 and the area including marker 2 are separated by a set of points at which the connection cost with marker 1 and the connection cost with marker 2 are equal. That is, the boundary of an area | region is connected by connecting the point where the connection cost with marker 1 and the connection cost with marker 2 become the same. The separation information inside and outside the pattern may take the form of connection cost with the marker of each label, not area information inside and outside the pattern as shown in FIG. 7.

In the area including the marker of each label, the connection cost with the marker is the same as that of the topographical distance from the marker by reconstruction by erosion of the terrain.

In FIG. 17, in the image represented by the solid line, it is determined that the lower the pixel value is, the more likely the inside of the area of the marker 1 is. In the image represented by the dotted line, it is determined that the lower the pixel value is, the more likely it is inside the area of the marker 2. Therefore, when the solid line corresponds to the image inside the pattern and the dotted line corresponds to the image outside the pattern, it can be determined from these images that it is likely to be internal and external.

Techniques for creating a contour by using the topographical distance from the marker as external energy and using an active contour are known. As dynamic contours, there are a snake method and a level set method in typical methods.

First, the snake method is demonstrated using FIG. 18A and 18B are diagrams showing the concept of the snake method. As shown in Fig. 18A, by arranging the point sequence around the shape for which the contour is to be obtained, and moving the point sequence to minimize the sum of the external energy and the internal energy, the contour is as shown in Fig. 18B. Obtain The external energy is the energy that is minimized when the dynamic outline (dotted line) is superimposed on the outline estimated from the image. Internal energy is the energy that decreases as the dynamic contour (splitting) satisfies the requirements (eg, small or short curvature) required as a natural edge. In addition, the initial value of a point sequence is the image which seems to be the inside of a pattern (for example, the image shown by the solid line of FIG. 17), and the image which seems to be outside the pattern (for example, the image shown by the dotted line of FIG. ) Is set so as to cover the inside of the pattern and not protrude out of the pattern.

Nonpatent literature 1 describes a method using a topographical distance from a marker as external energy and applying a snake.

Subsequently, the level set method will be described with reference to FIG. 19. In the level set method, the dynamic contour is not represented as a point sequence, but as the contour of the curved surface (subsidiary function) expanded one-dimensionally as shown in Fig. 19A. In the level set method, it is possible to easily express the change in the topology of the contour by modifying the auxiliary function. Non-patent document 2 describes a method using the topographical distance from a marker as external energy and applying the level set method.

Both the snake method and the level set method can use the connection cost with the marker instead of the topographical distance from the marker. The connection cost with the marker can be made in less time than the topographical distance from the marker.

In addition, the edge emphasis image can also be input to the outline preparation unit 105. In this way, an edge emphasis image can be used when creating external energy.

In the level set method, the change in the topology of the contour can be easily expressed by modifying the auxiliary function, so that the contour can be extracted satisfactorily even if there is a defect in the semiconductor pattern accompanied by a change in the topology such as necking or bridging. have.

(5) Theorem

In the present invention, the pattern data for specifying the inside and outside of the pattern (semiconductor pattern) formed on the target sample by thinning the design data for generating the inspection target sample (for example, a semiconductor wafer). (This is the initial zone). Then, referring to the pixel value (luminance value) of the edge-enhanced image of the target sample image, the region represented by the internal and external pattern information is expanded (growing the initial region by region), and the inner region and the outer region of the pattern are connected. Region division is performed by using a location as a boundary of a pattern to generate a pattern outline of a target sample. By doing in this way, the outline of a pattern which does not have the interruption | overload, the surplus, the overlap, the whisker, etc. in the middle of an outline can be obtained.

When generating the outline, the area indicated by the internal and external pattern information is expanded so that the area inside the pattern becomes the edge portion indicated by the edge emphasis image (the obtained edge emphasis image may be broken in the middle). Since it does not overlap%, area | region division is made so that it may overlap. For example, region division is performed based on the Marker Controlled Watershed Segmentation method. Alternatively, the pattern contour of the target sample is generated by performing dynamic contour extraction processing using the external energy representing the pattern inner region and the external energy representing the pattern outer region from the pattern internal and external prescribed information. Examples of this dynamic contour extraction processing include a snake method and a level set method. In the level set method or the snake method, the information separating the inside and the outside of the pattern includes the connection cost from the initial area or information equivalent thereto, and the SEM outline is created from the dynamic outline using this information. Thereby, a more natural outline can be obtained (especially in the case of the snake method). In addition, the level set method makes it possible to change the topology of the area separated by the SEM contour. Thereby, the contour of a semiconductor pattern can be obtained favorably even if defects, such as necking and bridging of a pattern, arise.

In addition, when the sample image (SEM image) is deformed to the extent that the latter cannot contain the former as compared with the design data, the design data is deformed so that the design data includes the target sample image, and the inside and outside of the pattern from the deformed design data. The regulation information may be generated. Thereby, even if the deformation of a semiconductor pattern is large, the outline of a semiconductor pattern can be obtained favorably.

The edge emphasis image may be created by combining the outputs of a plurality of filters with respect to the SEM image. The contour of the semiconductor pattern can be obtained satisfactorily also with respect to the SEM image with no white band.

1: Measurement / Inspection System
10: image processing unit
20: electron microscope body portion
30: SEM image generating unit
40: control calculator
50: input unit
60 display device

Claims (9)

An image processing apparatus for extracting an outline of a pattern formed on a sample to be measured or inspected,
A pattern internal and external regulation information generation unit which performs thinning processing on design data for generating a target sample to generate pattern internal and external regulation information defining inside and outside of the pattern formed on the target sample;
An edge emphasis unit for generating an edge emphasis image of the target sample image obtained by irradiating an electron beam to the target sample;
Contour generation section for performing region segmentation by expanding the region represented by the inside and outside pattern information while referring to the pixel value of the edge-enhanced image to generate a pattern contour of the target sample.
The image processing apparatus comprising: The method of claim 1,
And the contour generation unit expands an area represented by the inside and outside pattern definition information, and divides the area so that an area inside the pattern overlaps an edge portion shown in the edge emphasis image. 3. The method of claim 2,
And the contour generating unit divides the area based on the Marker Controlled Watershed Segmentation method. 3. The method of claim 2,
The contour generation unit generates a pattern contour of the target sample by performing dynamic contour extraction using the external energy representing the pattern inner region and the external energy representing the pattern outer region from the pattern internal and external prescribed information. Image processing apparatus. The method of claim 1,
The pattern internal and external definition information generation unit performs a peeling process on the image of the inner region and the outer region of the pattern in the design data with reference to the position alignment information acquired by matching the target sample image with the design data. And an inside and outside pattern defining information. The method of claim 1,
The pattern internal and external definition information generation unit deforms the design data so that the design data includes the target sample image by referring to the position alignment information acquired by matching the target sample image with the design data, and then deforms the design data. And the inside and outside pattern defining information are generated from the image processing apparatus. The method of claim 1,
And the edge emphasis unit generates a plurality of edge emphasis images by performing a plurality of filter processes on the sample image, and synthesizing respective filter processing images. An electron microscope device for generating an object sample image from information obtained by irradiating an electron beam to a sample to be measured or inspected;
The image processing apparatus of claim 1
Measurement / inspection system, characterized in that it comprises a. A computer-readable recording medium having recorded thereon a program for causing the computer to function as the image processing apparatus of claim 1 when executed by a computer.

Images